Pandas convert types and set invalid values as na
Is it possible to convert pandas series values to a specific type and set those n / a items that cannot be converted?
I found Series.astype(dtype, copy=True, raise_on_error=True)
with and installed raise_on_error=True
to avoid exceptions, but this will not set invalid elements to na ...
Update
More precisely, I want to specify the type to which the column should be converted. For a series containing values [123, 'abc', '2010-01-01', 1.3]
and type conversions to float
, I would expect [123.0, nan, nan, 1.3]
as a result, if selected datetime
, only the value series[2]
will contain a valid datetime value. convert_objects
doesn't allow for such flexibility, IMHO.
source to share
I think you might be lucky with convert_objects
:
In [11]: s = pd.Series(['1', '2', 'a'])
In [12]: s.astype(int, raise_on_error=False) # just returns s
Out[12]:
0 1
1 2
2 a
dtype: object
In [13]: s.convert_objects(convert_numeric=True)
Out[13]:
0 1
1 2
2 NaN
dtype: float64
Update: In more recent pandas, the method is convert_objects
deprecated.
In favor pd.to_numeric
:
In [21]: pd.to_numeric(s, errors='coerce')
Out[21]:
0 1.0
1 2.0
2 NaN
dtype: float64
It's not quite as powerful / magical as convert_objects
(which also worked on DataFrames), but works well in this case too. Read the section on converting objects in docs for other functions to_*
.
source to share